def __init__(self, name, container, application, processor_type):
self.application = application
self.container = container
self.name = name
print("processor_type", processor_type)
self.processor = util.cpu(processor_type)
self.mem = util.memory(processor_type)
def save_weights(self, savedfile, cutoff=0):
if cutoff <= 0:
cutoff = len(self.blocks) - 1
fp = open(savedfile, 'wb')
# Attach the header at the top of the file
self.header[3] = self.seen
header = self.header
header = header.numpy()
header.tofile(fp)
# Now, let us save the weights
for i in range(len(self.module_list)):
module_type = self.blocks[i + 1]["type"]
if (module_type) == "convolutional":
model = self.module_list[i]
try:
batch_normalize = int(self.blocks[i +
1]["batch_normalize"])
except:
batch_normalize = 0
conv = model[0]
if (batch_normalize):
bn = model[1]
# If the parameters are on GPU, convert them back to CPU
# We don't convert the parameter to GPU
# Instead. we copy the parameter and then convert it to CPU
# This is done as weight are need to be saved during training
cpu(bn.bias.data).numpy().tofile(fp)
cpu(bn.weight.data).numpy().tofile(fp)
cpu(bn.running_mean).numpy().tofile(fp)
cpu(bn.running_var).numpy().tofile(fp)
else:
cpu(conv.bias.data).numpy().tofile(fp)
# Let us save the weights for the Convolutional layers
cpu(conv.weight.data).numpy().tofile(fp)
def save_weights(self, savedfile, cutoff = 0):
if cutoff <= 0:
cutoff = len(self.blocks) - 1
fp = open(savedfile, 'wb')
self.header[3] = self.seen
header = self.header
header = header.numpy()
header.tofile(fp)
for i in range(len(self.module_list)):
module_type = self.blocks[i+1]["type"]
if (module_type) == "convolutional":
model = self.module_list[i]
try:
batch_normalize = int(self.blocks[i+1]["batch_normalize"])
except:
batch_normalize = 0
conv = model[0]
if (batch_normalize):
bn = model[1]
cpu(bn.bias.data).numpy().tofile(fp)
cpu(bn.weight.data).numpy().tofile(fp)
cpu(bn.running_mean).numpy().tofile(fp)
cpu(bn.running_var).numpy().tofile(fp)
else:
cpu(conv.bias.data).numpy().tofile(fp)
cpu(conv.weight.data).numpy().tofile(fp)
